BIL-722 ADVANCED TOPICS IN COMPUTER VISION a da Ba , N10266943 - - PowerPoint PPT Presentation
BIL-722 ADVANCED TOPICS IN COMPUTER VISION a da Ba , N10266943 Paper: Searching for objects driven by context Authors: Bogdan Alexe, Nicolas Heess, Yee Whye Teh, Vittorio Ferrari PURPOSE: OBJECT DETECTION Among many problems, all
Among many problems, all the methods exhaustively search the object
All the methods evaluates all the possible windows. This process is very slow and also unnatural.
Cognitive search shows that humans don’t do that. Instead search
Learn an object’s relative
An ideal search strategy would
1.
W1 is sky, cars occur below sky so look below.
2.
W2 is road, cars occur on the road, look just below the road
3.
There is a car part inside W3, look surrounding patches.
4.
W4 is a car.
1.
2.
1.
Similar position/appearance duo searched in the training set.
2.
Each of these similar patches votes for a new position.
3.
Method accumulates these votes as probability maps and decides where to look next.
3.
1.
If current window similar enough to a car, search is over.
A window is represented by these vector:
Window features 𝑧𝑢 consists of:
Normalized location and scale of the window HOG Histogram of the window Classifier score Displacement vector:
Intersection over union with the ground truth box Normalized Hamming distance to the ground truth box Absolute difference in the window classifier with the ground truth box
Extract uniformly distributed windows from all the training images,
For a test image: 1.
2.
3.
Calculate probability map with the new window in test image 𝑥𝑢
𝑥𝑢: Current window in test image. 𝑥𝑚: Window from training set.
Normalize each probability map and integrate all the past maps. Integrate all maps to form the overall probability map using
After 𝑈 iteration, output a single window which has highest classification
𝑢
This is a downside. Method assumes that there is only one instance in
There is a weight for each class in similarity kernel stage. This weights defines each patch’s importance for each object class.
An object classifier is trained for each class.
For each class, one root HOG filter and several part HOG filters are trained. Root and part filters summed with weights according to Felzenswab’s work.
For each class, training split is used for classifier learning.
Experiments conducted on PASCAL
A highly challenging dataset which contains 20 object classes witch bounding box
annotations.
Validation set is used for testing. Mean Average Precision over all classes and detection rate and number
Comparison of ? With Felzenszwalb et al. PAMI 2010
Experiments run on a Intel i7 processor powered PC. It can be seen that compared window count is significantly lower than
It is said that deformable part model approach takes 92s while
Pros:
Fast and logical search Can be applied with any classifier/feature
Cons:
Assumes only one instance exists. Dataset dependent?